Preamble

Simulations details:

  • 4 graph types: erdös, cluster, tree and scale-free
  • 2 global parameters: number of nodes (d), number of observations
  • Some additive parameters depending on graph types: densit and r for clusters, edge probability for erdös
  • Running on 100 generated graphs
  • Results are available with or without covariates
  • The covariates used are a set of 3 (cbind(rep(c(0,1),each=n/2),rnorm(n,8,0.5), round(runif(n)*10))) one binary, one gaussian and a categorical one with 10 groups. This last covariate is the hardest for SpiecEasi.

Each simulation (type x parameter x cov/no cov) takes about 100 minutes on my mac, parallellized on 3 cores.

AUC

Erdös

edge probabilty

d

n

Cluster

density

r

d

n

Tree

d

n

Scale-free

d

n

Add a new chunk by clicking the Insert Chunk button on the toolbar or by pressing Cmd+Option+I.

Precision Recall

Table

Data about precresion/recal criteria has been extracted from simulations under frames “precrec.auc”, in each files. Column names are not straightforward:

  • var is the value of the parameter (ex: 10 nodes)
  • param is the graph number
precrec <- as_tibble(readRDS("~/simulations/Simu/PLNcov/erdos/d/precrec.rds"))
precrec

Figures on all genrated graphs

Smoothed curves including results for all graphs on a partiuclar value of the tested parameter. Here : numerber of nodes fixed at 10.

Mosaic plot of Precision/Recal curves for different values of the parameter (here number of nodes is 10, 16, 20 or 30), summary results on all generated graphs.